Deep Learning to Improve the Sensitivity of Higgs Pair Searches in the 4b Channel at the LHC

The Higgs self-coupling is crucial for understanding the structure of the scalar potential and the mechanism of electroweak symmetry breaking. In this work, utilizing deep neural network based on Particle Transformer that relies on attention mechanism, we present a comprehensive analysis of the measurement of the trilinear Higgs self-coupling through the Higgs pair production with subsequent decay into four $b$-quarks ($HH\to b\bar{b}b\bar{b}$) at the LHC. The model processes full event-level information as input, bypassing explicit jet pairing and can serves as an event classifier. At HL-LHC, our approach constrains the $\kappa_\lambda$ to $(-0.53,6.01)$ at 68\% CL achieving over 40\% improvement in precision over conventional cut-based analyses. Comparison against alternative machine learning architectures also shows the outstanding performance of the Transformer-based model, which is mainly due to its ability to capture the correlations in the high-dimensional collision data with the help of attention mechanism. The result highlights the potential of attention-based networks in collider phenomenology.